Development of rapid and highly accurate method to measure concentration of fibers in atmosphere using artificial intelligence and scanning electron microscopy

Our plain-English summary of Development of rapid and highly accurate method to measure concentration of fibers in atmosphere using artificial intelligence and scanning electron microscopy, by Y. Iida, K. Watanabe, Y. Ominami, et al (Journal of Occupational Health, 2021).

The study

• Authors: Y. Iida, K. Watanabe, Y. Ominami, T. Toyoguchi, T. Murayama, M. Honda
• Journal: Journal of Occupational Health
• Published: 2021;63(1):e12238
• Cited by: 4 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Aim: We aimed to develop a measurement method that can count fibers rapidly by scanning electron microscopy equipped with an artificial intelligence image recognition system (AI-SEM), detecting thin fibers which cannot be observed by a conventional phase contrast microscopy (PCM) method.

Methods: We created a simulation sampling filter of airborne fibers using water-filtered chrysotile (white asbestos). A total of 108 images was taken of the samples at a 5 kV accelerating voltage with 10 000X magnification scanning electron microscopy (SEM). Each of three expert analysts counted 108 images and created a model answer for fibers. We trained the artificial intelligence (AI) using 25 of the 108 images. After the training, the AI counted fibers in 108 images again.

Results: There was a 12.1% difference between the AI counting results and the model answer. At 10 000X magnification, AI-SEM can detect 87.9% of fibers with a diameter of 0.06-3 μm, which is similar to a skilled analyst. Fibers with a diameter of 0.2 μm or less cannot be confirmed by phase-contrast microscopy (PCM). When observing the same area in 300 images with 1500X magnification SEM-as listed in the Asbestos Monitoring Manual (Ministry of the Environment)-with 10 000X SEM, the expected analysis time required for the trained AI is 5 h, whereas the expected time required for observation by an analyst is 251 h.

Conclusion: The AI-SEM can count thin fibers with higher accuracy and more quickly than conventional methods by PCM and SEM.

© 2021 The Authors. Journal of Occupational Health published by John Wiley & Sons Australia, Ltd on behalf of The Japan Society for Occupational Health.

In plain English

This laboratory study tested whether a scanning electron microscope paired with artificial intelligence (AI) image-recognition software could count asbestos-type fibers faster than people. The authors made a simulated air-sampling filter using chrysotile (white asbestos) from three production areas, mixed into water and filtered. They took 108 microscope images at 10,000 times magnification. Three expert analysts counted the fibers to create a model answer. The software was trained on 25 of the images and then tested on all 108. Of 132 fibers, the AI detected 87.9% and missed 12.1%, mostly thin fibers. Its false detection rate was 21.4%, mainly because it counted fibers of the filter material itself. The electron microscope showed many thin fibers that the standard phase-contrast optical method could not count. The AI took 0.03 minutes per image, compared with 1.4 minutes for an analyst. The authors concluded the AI detected fibers 50 times faster than humans, but said accuracy needs improvement and the method needs further validation. They noted it was not tested on amosite, crocidolite, tremolite or other fibers or dust, and that the training set was small.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Laboratory method-development study comparing fiber counts made by human analysts and by AI image-recognition software on scanning electron microscope images of a simulated chrysotile filter
• Who was studied: No human subjects. The study used 108 scanning electron microscope images (10,000X magnification) of a laboratory-made chrysotile filter. Each image was counted by three expert analysts, and 25 images were used to train the AI.
• Exposure: Chrysotile from three production areas (listed as Canada B/L 4T-500, Soviet Union or Russian M-6-40, and Brazilian 4T), dispersed in water and filtered through a methylcellulose membrane filter to simulate an air sample. Fibers 0.06 to 3 micrometers wide were analyzed.
• Main results:
  • Of 132 total fibers, the AI detected 87.9%. Its false negative rate was 12.1%: 16 fibers were found by analysts but missed by the AI, all of them thin fibers.
  • The AI's false detection rate was 21.4%. It detected 36 particles that should not be counted as fibers, of which 32 were fibers of the filter itself.
  • The main cause of false detections was the AI counting fibers of the sampling filter as fibers.
  • Counted on the same filter, the SEM at 10,000X found many thin fibers that phase-contrast microscopy could not count. Estimated fibers per phase-contrast field were 6 by phase-contrast microscopy, 42 and 58 by SEM at 1500X (two analysts), and 401 to 471 by SEM at 10,000X (three analysts).
  • An analyst needed 1.4 minutes per image at 10,000X. The AI needed 0.03 minutes per image.
  • For an area equal to the 300 images at 1500X specified in the Asbestos Monitoring Manual, viewed at 10,000X, the expected time was 5.4 hours for the trained AI and 251 hours for an analyst.
• The authors' conclusion: The authors concluded that the AI could detect fibers 50 times faster than humans. They also said that, given the 12.1% false negative and 21.4% false detection rates, accuracy needs improvement, and that the method must be further verified and validated.
• Limitations the authors state:
  • Only a simulated filter made from water-filtered chrysotile was tested. The method was not verified for amosite, crocidolite, tremolite, or other fibers or dust.
  • Only 25 images were used to train the AI, which the authors described as not a large number.
  • The methylcellulose filter's sponge-like structure looked like fibers, which led the AI to count filter fibers as fibers.
  • The study did not analyze what the fibers were made of.
  • The authors stated the results are not sufficient and further verification and validation are needed.

Topics

Medical terms in this study, explained in our Lexicon: Scanning Electron Microscopy (SEM), phase contrast microscopy (PCM), Chrysotile.

Related studies

• Rapid fiber-detection technique by artificial intelligence in phase-contrast microscope images of simulated atmospheric samples — T. Yamamoto, K. Iwasaki, et al. (2024)
• A technical comparison of evaluating asbestos concentration by phase-contrast microscopy (PCM), scanning electron microscopy (SEM), and analytical transmission electron microscopy (ATEM) as illustrated from data generated from a case report — R.F. Dodson, S.P. Hammar, et al. (2008)
• Artificial intelligence-based diagnosis of asbestosis: analysis of a database with applicants for asbestosis state aid — K.B.W. Groot Lipman, C.J. de Gooijer, et al. (2023)
• Study of inorganic particles, fibers, and asbestos bodies by variable pressure scanning electron microscopy with annexed energy dispersive spectroscopy and micro-Raman spectroscopy in thin sections of lung and pleural plaque — C. Rinaudo, A. Croce, et al. (2010)
• A total sample preparation method for the measurement of airborne asbestos and other fibers by optical and electron microscopy — N. Kohyama, S. Kurimori (1996)
• Environmental scanning electron microscopy technique to identify asbestos phases inside ferruginous bodies — A. Croce, M. Musa, et al. (2013)

Browse more research in Journal of Occupational Health, or return to medical research.